Search:The WebTripod

From Ethan Zuckerman, Vice President for Business and Research Development:

As if working with the bright folks at Tripod isn't humbling enough, I've once again taken up computer chess. My dear overworked Macintosh is a patient enough opponent, and I'm proud to announce that I actually won a game this morning, making my win-loss ratio about as good as that of this year's Chicago Cubs.

This is what amazes me about the upcoming Kasparov/Deep Blue rematch — human beings still beat computers at this game.

Let me back up for a moment for those of you who aren't checking IBM's Web site every day, counting down to May 3rd. Deep Blue is a large, highly-parallel, extremely well-programmed computer designed — solely — to play chess. Gary Kasparov is a fierce, flamboyant, articulate, slightly built 34-year old Russian who's held the world chess championship for the past 12 years. The two faced off last February, and Deep Blue shocked the world by beating Kasparov in the first game. Kasparov drew two and won three of the remaining five games, drawing on an uncanny ability to switch playing styles mid game, a strategy that evidently befuddles Deep Blue as much as it frustrates Kasparov's human opponents.

The two face off again this year, from May 3 to May 11 at The Equitable Center in New York. The Deep Blue team (five programmers, one grandmaster, and 1.4 tons of IBM-sponsored circuitry) have been hard at work trying to ready the machine for the match. Though I haven't been able to find a Vegas bookmaker willing to lay odds on the event, many knowledgeable commentators (and I) feel that Kasparov will emerge victorious again this year. (My Mac crashed three times as I typed that last sentence. I think I know where my machine's loyalties lie.)

A comparison of the opponents could sway most odds makers to IBM's camp. Deep Blue, wired with numerous specially-designed "chess processors" wired in parallel, is able to consider 200,000,000 chess positions per second. Gary Kasparov, with roughly the same hardware as you and me, is able to consider three positions a second. (In case you're curious, the Mac which keeps whupping my ass considers about 5000 positions a second.) Deep Blue has access to a vast library of grandmaster games programmed into its memory. Kasparov has access to a vast library of chess knowledge, but does not have the luxury of perfect digital recall. Deep Blue never gets tired, flustered, hungry or overwhelmed. Kasparov does — and moreover, Kasparov's techniques for psyching out his opponents have no effect on Deep Blue.

Yet Kasparov won last year — decisively. After the shock of the first match, Kasparov came back strong, and in the last two games of the match, most commentators feel that he made Deep Blue look bad.

All of which is an interesting commentary on the difference between human and machine intelligence. There are very few tasks in which human beings and computers compete. Things that computers are good at — remembering huge quantities of data, doing arithmetic really fast, evaluating thousands of possibilities in a few seconds — are things human beings are poor at. Yet very few of us feel like we need to compete with our computers.

Similarly, things that humans are good at — producing speech, understanding the meaning of text, drawing conclusions from data given — computers are terribly bad at. You only need to look at how far computer science is from the dream of HAL to realize just how short we've fallen. Stanley Kubrick, who created one of the most scientifically accurate films ever in 2001, seriously believed that computers with the capabilities of HAL (human-like speech, accurate visual input, free will, understanding) could be possible by the late 1990s. Instead, we're still trying to get the damn things to beat us at chess.

Humans are ignoring machines — we have an uncanny ability to edit out unimportant information and focus on the data we need. This, incidently, is probably how Kasparov is able to consider 199,999,997 positions a second fewer than Kasparov and still win games. The three positions he considers are relevant, meaningful, likely, and he's merely picking the best of them. Many of Deep Blue's possibilities are likely to be crap, and the machine needs to evaluate and discard each of them independently.

Computers are capable of playing "perfect" checkers — they can evaluate all possible games after you've made your first move, and follow through on a strategy that is guaranteed to win. That's because there is a finite and relatively small number of checkers games, and a powerful enough computer can be aware of all those possible games. No amount of human intuition and creativity can triumph over a computer that has "solved" the game.

Chess is too difficult to solve (currently) by brute force — by one calculation, there are about 10^44 (ten to the forty fourth power) possible chess games, a number which is significantly larger than the number of atoms estimated to compose the universe. This number, evidently, is small enough that it matters whether Deep Blue can consider 200 million a second rather than 100 million a second, but not small enough that it can "solve" the game. Instead, Deep Blue has had to learn a lot of the same techniques Kasparov uses to be a great player — memorizing common openings, playing thousands of endgames, learning from important historical games. It's an amazing tribute that a game invented, at the latest, in the 11th century, remains such an amazing and balanced test of different forms of intelligence today.

So, for the record, my money's on Kasparov, although I won't shed any tears if Deep Blue emerges victorious. Until computers are capable of creating a game as beautiful, subtle and challenging as chess, I'm not too worried about the value of human intelligence.

And I just realized that if I ripped half the RAM out of this Mac, I might have a better shot with tomorrow's games. Where's that screwdriver?

-E

---

Read more "Letters from Tripod" in the archive.

---